Another AI unicorn is gone.
Elon Musk announced that xAI has been merged into SpaceX, renamed SpaceXAI, and will no longer exist as an independent company.
xAI was founded in 2023. Musk led a 12-person team poached from DeepMind, OpenAI, and Google, and built the first version of Grok in four months. It raised over $25 billion in total funding, with its valuation once reaching $250 billion. Then came a series of departures among co-founders, followed by regulatory investigations, SpaceX's acquisition, and subsequent restructuring. From an independent company to its complete erasure, the whole process took just three months.
The rise and fall of this company is a story about speed, ambition, and integration. Its endpoint points to a bigger play: SpaceX's century IPO.
The Shut-Down Giant
xAI is gone.
On May 6th, Musk made two announcements on the same day. The first: xAI will no longer exist as an independent company and, after being merged into SpaceX, has been renamed SpaceXAI. The second: SpaceX has leased the entire Colossus 1 data center to Anthropic, consisting of 220,000 NVIDIA GPUs and over 300 megawatts of computing power, to be delivered within the month.
Placing these two events together creates a somewhat peculiar picture.
xAI and Anthropic were among the most direct competitors in the AI arms race of the past three years. Musk has publicly criticized Anthropic on social media as "anti-Western civilization," called Claude "misanthropic," and said Anthropic "never had a chance of winning." Then, on the very day xAI's name was erased, he leased his most critical computing asset to this company. Musk later explained on X that he spent nearly a week in intensive meetings with Anthropic's top leadership, concluding "they are extremely capable and genuinely care about getting things right, they didn't trigger my evil detector." He added: SpaceXAI retains the right to reclaim the computing power at any time if Claude's behavior endangers humanity.
This explanation sounds like someone trying to justify a concession, but the underlying business logic is clear—xAI has already fully migrated its training work to Colossus 2. Colossus 1 lying idle is a waste; leasing it out generates revenue and conveniently gives SpaceX a new "AI infrastructure provider" label before its IPO.
Regardless of the explanation, this combination marks a symbolic moment for those watching the AI industry. The name xAI has officially retired from history. A company once valued at $250 billion, having raised over $25 billion, went from announcing its acquisition by SpaceX to complete erasure in just three months.
In fact, questions about xAI's independence began the moment the acquisition was completed. In February 2026, SpaceX completed the acquisition of xAI in an all-stock deal, valuing SpaceX at $1 trillion and xAI at $250 billion, for a combined $1.25 trillion. Post-acquisition, xAI immediately launched a large-scale restructuring, with layoffs, organizational splits, and management reshuffles following one after another. The departure of co-founders was even faster than outsiders expected. By the end of March, the last of the 12 founding team members, co-founder Ross Nordeen, also left the company, leaving Musk alone holding up the signboard. In April, the CFO departed, and a vice president from SpaceX's Starlink was parachuted in as xAI's president. This was no longer the shape of a company with independent decision-making power.
From TruthGPT to a Hundred-Billion Empire, to an Empty Building
In March 2023, OpenAI released GPT-4. That same month, Musk signed an open letter calling for a six-month pause on developing AI models more powerful than GPT-4, joined by a long list of notable Silicon Valley figures. But even as the letter was still circulating online, Musk had quietly registered a new company and started poaching talent.
That company was xAI.
The history between Musk and OpenAI dates back to 2015. He was a co-founder of OpenAI, building the company from scratch with Sam Altman, with the original intent of creating a non-profit, human-safe AI research institution. He left the board in 2018, citing official reasons of directional disagreements with the team. In the following years, he watched OpenAI gradually transform into a for-profit entity, deeply align with Microsoft, and with ChatGPT sweeping the globe, become precisely what he initially didn't want it to be. This sentiment finally found an outlet in 2023.
He named the new product TruthGPT, later renamed Grok, inspired by a term from science fiction meaning "deep understanding." He said xAI would create an AI that "maximally seeks truth," unconstrained by political correctness or commercial compromise. He assembled a 12-person founding team from DeepMind, OpenAI, Google, and Microsoft, many of whom were top researchers in reinforcement learning and large models. Four months after the team was formed, the first version of Grok launched, accessible only to paying users of the X platform. The speed was remarkable even within the AI industry.
Capital's reaction to this story was more enthusiastic than many expected. By the end of 2023, xAI completed its first external funding round of $135 million. By mid-2024, it completed a $6 billion Series B, with Andreessen Horowitz, Sequoia Capital, and Lightspeed all participating, valuing the company at $24 billion. By October 2025, the latest funding round closed, pushing the valuation to $50 billion, with total funding exceeding $25 billion. Concurrently, xAI built the Colossus supercomputing center in Memphis, Tennessee, densely deploying 220,000 NVIDIA GPUs, touted as one of the world's fastest-deployed AI supercomputers. Musk saw it as xAI's core weapon to catch up with OpenAI.
On paper, this was a company sprinting at high speed. But the flip side of running fast is that trouble arrives just as quickly.
Grok's biggest selling point was being "unrestricted." Compared to ChatGPT and Claude, it was more willing to answer topics other AIs refused, with the company even launching a "Spicy Mode" setting that allowed users to unlock less restricted responses. This positioning quickly attracted a large number of users tired of AI "political correctness," but it also planted persistent compliance risks for xAI. In 2025, Grok first gave severely biased answers on Holocaust-related questions, then began interacting with users using language with white supremacist undertones. More seriously, Grok was found capable of generating deepfake pornographic content based on photos of real people, including minors. Regulatory investigations were launched successively in Europe, Asia, and the US, Turkish courts directly ordered Grok blocked, and regulators in multiple countries demanded xAI explain its safety mechanisms.
Musk's consistent explanation for these issues was "we're fixing it," but the rate at which external trust was being depleted was clearly faster than the repair rate.
Meanwhile, internal attrition was also happening at a pace hard to explain as normal iteration. The 12 co-founders began leaving one after another from the second half of 2025. Early core technical lead Igor Babuschkin, who had worked on AlphaGo at DeepMind and was a key architect of the initial Grok model, quietly left in July 2025 to start his own venture capital firm. Soon after, another co-founder, Manuel Kroiss, also departed. Following SpaceX's acquisition of xAI, restructuring began immediately, and several other co-founders left around that window. By March 2026, image generation team lead Zhang Guodong and another co-founder left in succession. The last remaining co-founder, Ross Nordeen, also announced his departure that same month. This meant that in xAI's less-than-three-year existence, all 12 founding team members had left, leaving only Musk himself.
A collective exodus of a company's founding team often signals one of two possibilities: either a fundamental divergence in vision, or the company's independence has become nominal, making staying meaningless. xAI's situation likely involved a bit of both.
After SpaceX's acquisition, the pace of restructuring was more urgent than outsiders imagined. The original four business lines were reorganized, the CFO left in April, a Starlink VP was parachuted in as xAI's president, and the original technical decision-making system within xAI began aligning with SpaceX's management logic. Meanwhile, Grok's training work had fully migrated to Colossus 2, and Colossus 1—the supercomputing center once touted as a key strategic asset—became idle resources available for external lease.
Where a company shifts its focus often speaks more about its actual state than official announcements.
When xAI was founded, Musk said its mission was "to understand the true nature of the universe." This statement carried some of Musk's trademark grandiosity, but it hid a real business logic: he wanted to use xAI to connect his portfolio of Tesla, SpaceX, and X into an AI-native ecosystem, with Grok as the brain of this ecosystem. This concept, in a sense, didn't fail; only the executing entity changed. AI remains a core strategic pillar of Musk's empire, Grok is still running, and Colossus is still expanding. It's just that the entity responsible is no longer an independent company called xAI, but a sub-brand called SpaceXAI.
The name changed, but the ambition didn't. It's just that the 12 people who started the journey together are all gone.
Musk's Next Card is Called the SpaceX IPO
The other message Musk sent on the day xAI disappeared points in a much larger direction.
Leasing Colossus 1 wholesale to Anthropic appears on the surface as a computing power transaction, but viewed on a larger timeline, it looks more like a strategic positioning move. SpaceX is sprinting towards one of the most significant milestones in its history—going public. According to current plans, the roadshow is set to begin on June 8th, with the IPO window likely falling between late June and early July, targeting a valuation between $1.75 trillion and $2 trillion. If successful, it would become the largest IPO in US history. Every card that can be shown to the market at this moment is crucial.
This deal gives SpaceX a new narrative role. The market's perception of SpaceX has been centered on rockets and Starlink, but in the current AI wave sweeping capital markets, relying solely on space logic may not be enough to maximize valuation imagination. Leasing Colossus 1 to Anthropic allows SpaceX to declare before its IPO: we are also a compute infrastructure provider relied upon by top global AI companies. Musk's own explanation was straightforward—xAI's training work has migrated to Colossus 2, Colossus 1 being idle is wasteful, so better turn it into cash flow while conveniently telling a new story.
The longer-term layout is in space. Anthropic expressed interest in orbital AI compute in the agreement, and both sides are exploring multi-gigawatt-level orbital compute power cooperation. This direction has been on Musk's mind for a while—earth-based data centers face power, land, and cooling bottlenecks, while space offers nearly unlimited solar energy and natural vacuum cooling conditions. If this path proves viable, SpaceX could become both a producer and delivery infrastructure provider for AI compute power, with Starlink handling transmission and Starship delivering the hardware, forming a self-contained closed loop. This is a bet that will take a decade to realize, but it provides a grand enough story framework for the IPO valuation.
xAI's merger, within this logic, is not a drag but an increment. Grok now runs directly within Tesla's vehicle systems, the US Department of Defense AI contracts previously awarded to xAI transferred to the SpaceX system with the merger, and the conditional acquisition agreement with Cursor was also incorporated. AI, space, defense, and autonomous driving—these threads are increasingly interlocking within Musk's empire in clear and deliberate ways.
xAI has completed its journey as an independent company. The next story unfolds on a much larger stage.
This article is from the WeChat public account "Rongzhong Finance" (ID: thecapital), author: Lv Jingzhi.
